There's an essay making the rounds right now - Matt Shumer's "Something Big Is Happening" - and it's been viewed 80+ million times for a reason (for what it's worth... it was at 6 million views when I read it a few days ago and I thought THAT was a lot. Now it's at 80 million+).
His argument is straightforward: AI has crossed a threshold, and most knowledge workers are about to feel it. He compares it to February 2020, when a few people were talking about a virus and most people thought it was overblown. Three weeks later the world changed.
I know exactly what that felt like, and not as a metaphor.
In February 2020, my husband and I were putting the finishing touches on the purchase agreement for a regional guide company in southern Utah. A highly leveraged transaction, big bet type move, the kind of deal where you either grow into it or it will likely bury you. During the due diligence, I was leaning hard on the skills I had learned in a previous life as a Harvard MBA and Manager at Bain Consulting, and we thought that we walked into the deal with eyes wide open. We knew the bet that we were making, and the seller knew it too.
Yet... when we closed the deal on March 6, 2020, taking on a massive amount of personally guaranteed debt, we closed out the day that should have been a celebration with a vague feeling of dread and, truthfully, zero understanding of what was actually on the horizon. Ten days later we had to shut the company down. We hadn't even finished onboarding and we were staring at what looked like bankruptcy before we'd started.
(Spoiler alert: we figured it out, pulling through - and we've since tripled revenue. But it was a WILD ride in those early months.)
That vague feeling of dread - the one where you know something is shifting but you can't see the shape of it yet - I think that's where a lot of people are with AI right now. And Shumer is right that most people are underestimating what's coming.
But that's not where I am. Not because I understand the big picture - none of us do - but because I DO understand the day-to-day leverage piece. I've spent months learning how to make these tools actually work inside my businesses, and I now have real evidence that AI can help me triple revenue again if that's what I decide to lean into.
I also noticed something in the responses to Shumer's essay: a lot of people read it, felt the urgency, opened ChatGPT or Claude, typed a question, got a decent-but-generic answer, and thought that's it?
That was me six months ago. But now... I spend every team meeting for each of my businesses every week talking about AI, new tools that I've built in the prior seven days, and new uses that my team members are discovering (or wishing for). Every week.
So here's what I actually did differently - the specific, unglamorous stuff - that turned AI from a mildly useful search engine into something that fundamentally changed the way I run my businesses and my life; something that is tangibly improving my quality of life on a daily basis.
1. I stopped chatting and started building.
Most people use AI in one-off chats. Ask a question, get an answer, close the tab. That's using maybe 10% of what's available.
I work in projects. In Claude (which is what I use - the Max plan at $200/month, and yes, it's worth it), a project is a persistent workspace where context accumulates. The AI remembers the project's purpose, the documents you've uploaded, the decisions you've made. A chat is a Post-it note. A project is an office.
Every time I start a new project, the first thing I do is tell Claude what the project is for and ask it to generate a set of highly-leveraged instructions for itself. I copy those instructions into the project. Now Claude isn't starting from zero every time I open a new conversation within that project - it has orientation, context, and constraints. Just like a human team member would.
Your Action: Whichever AI you decide to use - upgrade to the highest available level right now; you don't have to lock yourself into that decision, but give yourself that "gift" for a month to really wrap your head around what AI can do. If, after that month, you still think AI is marginally useful.... cancel your subscription.
And: stop working in individual, non-memory-accumulating chats.
2. I gave it a PhD on me.
One of the first things I did was ask Claude to get as smart on me as it could using all available online information - with PhD-thesis-level rigor. It produced a comprehensive longitudinal analysis of my life and career, and a detailed style guide for how I write and communicate.
Yes, I asked an AI to study me like a dissertation topic. The Bain in me couldn't help it. (The more of a digital footprint you have - a website, published writing, earned media, etc etc - the more useful this part will be.) The documents it produced were fascinating, and gave me words for patterns and anchors in my life that I didn't even know were there, but that I recognized the moment AI put them in front of me.
Those 'dissertation on Sunny' documents now live in every project where they're relevant. When Claude drafts something for me now, it's not generating generic corporate prose. It knows that I use spaced hyphens, that I hate inspirational platitudes, that my humor deflates rather than emphasizes, that I compress credentials into one sentence and move on. And, even more importantly, it knows the throughline of my work - the thread that connects Bain to endurance racing to running outdoor businesses to gender equity advocacy - because I made it do the homework.
Your Action: Give your AI foundational homework. Tell it who you are, where you work, what matters to you. Then tell it to get as smart on you / your work / your life as it possibly can, with PhD-level rigor and deep research. Make sure you iterate on the output of this, and that it flows into your AI's memory!
3. I told it exactly how to think.
Under Claude's settings, there's a personal preferences section. Most people leave it blank. Mine says this:
You are my critical-thinking thought partner. Optimize for correctness, clarity, and decision-quality over tone. Be terse. No filler, no praise, no "great question." Challenge assumptions. Identify missing info, edge cases, failure modes. If my plan is weak, say so directly and propose a better alternative.
That's the abridged version. The full thing is detailed and specific - output format, depth control, when to ask questions versus when to just answer. It fundamentally changes the character of every interaction. Without it, you're talking to a polite generalist. With it, you're talking to a partner who knows how you think and what you need.
Your Action: Think about all the things that may have annoyed you in past surface-level ChatGPT (or other AI) interactions. Feed that list to AI, and tell it to write a set of preferences or instructions that steer clear of those mannerisms.
4. I taught it to write like me (and to check its own work).
Claude has a feature called Skills - essentially, backend instructions that govern how it handles specific types of tasks. I asked Claude to build a detailed Skill around writing in my voice, using the style guide it had already created. But I didn't stop there. The Skill includes a feedback loop: Claude drafts something, then cross-checks the draft against my voice guide, identifies gaps, rewrites to close those gaps, and only then shows me the result. It asks if I want it to run the loop again.
The output isn't perfect. But it's close enough that I'm lightly editing rather than rewriting, which is a fundamentally different (and faster) workflow.
Case in point: this article was an AI-assisted production that took me about 5 minutes (OK, actually: 15 minutes) to create. Two minutes to prompt, five minutes to read the AI draft, three minutes for light edits, five minutes to upload and design on the website. And I'm really proud of the piece. It's 100% me: what, and how, I would write things manually.... just compressed into a 15 minute production timeline.
Your Action: Give your AI 4-5 long-form content examples that are genuinely YOU. Push it to get deeply insightful and analytical on distilling what makes your writing yours, then make it use that insight every time you ask it to write something.
Not quite sure how to actually do this? Take the above paragraph, paste it into your AI, and tell it: I don't know how to do this! Help?
5. I pushed back when it was wrong.
This one matters more than any setup tip. When I first asked Claude to help me build application scoring workbooks for a scholarship fund I manage, it walked me through a process that technically worked but required a ton of manual steps. It was a huge upgrade compared to the previous, non-AI powered process, but it still required a lot of human intervention, attention, and multi-step prompting.
So I told it: this process is too manual. Is there a faster way?
Claude's answer: "Yes, of course - just do XYZ."
The better solution was sitting right there the whole time. I just hadn't asked the right question. (If you're not occasionally wanting to throw your laptop at the wall over this, you're not pushing hard enough.)
This is the pattern I see over and over. AI will give you a B+ answer to a B+ question. If you want the A+ answer, you have to push. You have to say "this isn't good enough" and tell it why. You have to treat it like an infinitely patient, incredibly fast new hire who needs direction but will deliver if you give it enough information and specificity to work with. If you don't tell it the why, the what, and the how - just like a human employee - it's not going to be effective.
Your action: When AI isn't working for you, don't give up. Instead, tell AI "this isn't working for me. Get creative and give me other options / ideas / solutions to make it work for me."
6. I gave it every permission and all the context.
Web search, memory, file creation, past chat references - I turned everything on. The more information Claude has access to, the better its outputs. This sounds obvious but a surprising number of people are using these tools with half the capabilities toggled off because they never went into settings.
Your action: Go into settings and customize your permission settings. An AI that can't search the web and open websites for you in realtime is largely useless for most of us.
So is "something big" happening?
I don't know. Maybe. Probably. The tech people will tell you it's a revolution and the skeptics will tell you it's a bubble and I've spent enough time in both strategy consulting and ultramarathons to know that the people who are most confident about what's coming next are usually wrong about the timeline.
What I do know is this: I run four organizations, I'm training for a 350-mile foot race across Alaska, and my work is more fun than it's been in years. Not because AI made everything easy. Because it made the tedious parts fast enough that I get to spend my time on the parts that actually require me - the judgment calls, the relationship-building, the creative work, the stuff that happens when you're standing in a slot canyon with a client and not sitting at a laptop reformatting a spreadsheet.
The gap between people using AI casually and people who've spent a few hours configuring it properly is enormous. None of what I described above is hard. But none of it is default, either.